Neural network model lithium ion battery state prediction method based on convolution and bidirectional long and short time memory

Abstract

The invention belongs to the field of computer systems based on calculation models, and discloses a neural network model lithium ion battery state prediction method based on convolution and bidirectional long and short time memory, comprising the following steps: S1, data acquisition and data preprocessing; the method comprises the following steps: S1, constructing battery discharge data under a plurality of temperatures and different driving cycles by adopting a sliding window technology, S3, constructing a neural network model, namely a CNN-BiLSTM model, of a convolution and bidirectional long-short term memory network, S4, carrying out model migration, and S5, carrying out model migration according to an obtained CNN-BiLSTM-TL network model of a target domain. The SOC and the SOE can be accurately estimated, the charge and discharge control strategy of the battery can be optimized, the service life of the battery is prolonged, the efficiency of the battery is improved, and meanwhile the method is a key for guaranteeing efficient and stable operation of a battery system.

Description

Technical Field

[0001] The invention belongs to the field of lithium-ion battery state prediction, and specifically relates to a lithium-ion battery state prediction method based on a neural network model of convolution and bidirectional long short-term memory. Background Art

[0002] Lithium-ion batteries have the advantages of long cycle life, good charge and discharge performance, and easy portability. They are widely used in electric vehicles, unmanned boats, and electric boats, and have become the main power source for electric vehicles. In actual application, lithium batteries, as the main energy storage components and power sources of electric vehicles, directly affect the performance and safety of car driving. In actual use, batteries have unsafe behaviors such as overcharging and over-discharging, which may lead to frequent safety accidents such as fire and explosion. Moreover, factors such as long charging time, short battery life, and low estimation accuracy also restrict the development of electric vehicle technology. In addition, due to the strong coupling and nonlinearity of lithium battery systems, SOC (battery state of charge, indicating the ratio of the remaining power or charged power to the full power of the battery) and SOE (the ratio between the current energy stored in the battery and its maximum stored energy) are two important monitoring parameters in BMS (battery management system), which are two key parameters to describe the battery state. They directly affect the battery's charge and discharge performance and reliability. Improving the estimation accuracy of BMS for various battery state parameters has become a key issue in improving the safe and stable operation of new energy vehicles.

[0003] Convolutional Neural Network (CNN) is a feedforward neural network that is further improved and developed under the framework of BP neural network. Its design is inspired by the theory of visual receptive field and is widely used in multidimensional data processing, including time series, natural language, image recognition and other fields. In a fully connected neural network, each neuron receives the output from all neurons in the previous layer, so that a close relationship structure is formed between neurons, which helps to propagate information in the network and learn complex feature representations. CNN uses the principle of parameter sharing to reduce the number of parameters that the model needs to learn, thereby reducing the complexity of the model and enhancing the generalization performance of the model. A prominent feature of CNN is that in each layer of the network, each neuron node is only connected to some nodes in the previous layer, not all nodes, which is called local receptive field. A typical CNN is usually composed of convolutional layers, pooling layers and fully connected layers, and different types of layers have different functions.

[0004] Bidirectional long short-term memory neural network (BiLSTM) is also a recursive neural network model. It is improved based on the traditional long short-term memory neural network (LSTM). This model is widely used in long time series data. The BiLSTM network structure is composed of two LSTMs with different propagation directions, which can well mine the intrinsic connection between the current moment data and the past and future data. The BiLSTM network structure contains two independent LSTM layers. The functions of the two LSTM layers are to process the input time series data in the forward direction and the reverse direction. In the forward LSTM, the input data is sequentially input along the time axis, and the output of each moment is used as the input of the next moment. In the reverse LSTM, the input time series data is sequentially input in the opposite direction of the time axis, and the output of each moment is used as the input of the next moment as in the forward direction. Then the forward and reverse outputs of the LSTM are fused to obtain the final output of the BiLSTM network.

[0005] Transfer learning (TL) shows good applicability when dealing with insufficient target domain data and large source domain data. Moreover, this method can learn common information related to target domain SOC and SOE estimation from a large amount of source domain data. Among this information, only knowledge that is useful for target domain SOC and SOE estimation can be effectively transferred. Ultimately, this common knowledge is used to achieve accurate SOC and SOE estimation. Summary of the invention

[0006] The present invention provides a method for predicting the state of lithium-ion batteries based on a convolution and bidirectional long short-term memory neural network model. The method is capable of predicting the state of charge of the battery, automatically extracting the characteristic parameters of the battery dynamics, mining the relationship between the characteristic parameters, and establishing a high-precision prediction model; the convolution and bidirectional long short-term memory neural network model is used to quickly process high-dimensional parameters and verify the robustness and generalization of the model. The present invention can accurately estimate SOC and SOE, optimize the battery charge and discharge control strategy, improve the battery life and efficiency, and is the key to ensuring the efficient and stable operation of the battery system.

[0007] To achieve the above object, the present invention provides the following technical solution: a method for predicting the state of a lithium-ion battery based on a neural network model of convolution and bidirectional long short-term memory, the method comprising the following steps: S1: Data acquisition and data preprocessing, obtaining lithium-ion battery data, including battery discharge current, voltage, SOC and SOE variables; S2: Use sliding window technology to construct battery discharge data under multiple temperatures and different driving cycles to form matrix data suitable for input into CNN, and normalize the data to obtain data training set and data test set; S3: Build a convolutional and bidirectional long short-term memory neural network model, namely the CNN-BiLSTM model, input the data training set into the neural network model for training, so that the neural network learns the characteristics between the data; input the data test set into the neural network model to predict the battery state of charge and health state. When the model does not achieve ideal results when testing the data test set, adjust the parameters of the neural network model until the test results are ideal; S4: Model migration: freeze the network layers of the pre-trained CNN-BiLSTM model layer by layer manually. After freezing some network layers in the CNN-BiLSTM model, fine-tune the parameters of the unfrozen network layers according to the discharge data of different types of batteries, and build a joint estimation model of SOC and SOE for different types of batteries and different discharge intervals. The neural network model of convolution and bidirectional long short-term memory network transfer learning, namely, the CNN-BiLSTM-TL network model, is obtained. S5: According to the obtained target domain CNN-BiLSTM-TL network model, the target domain data is input into the model to estimate the battery state of charge and health state. During the experiment, the CNN-BiLSTM-TL network model uses the same small sample training set as in the fine-tuning process to estimate the battery's SOC and SOE.

[0008] Preferably, the specific method of step S1 is: according to the urban road cycle conditions, the voltage and energy change trends during the discharge process of the power battery tested at different temperatures, the discharge current, battery cell temperature, voltage, and SOC parameter model input parameters in the data set are extracted, and the battery state of charge and energy state are predicted as target data.

[0009] Preferably, the specific method of step S2 is: S2.1: Convert the data into a time series, determine the sliding time window length L to be 10 to reconstruct the data. If the current cycle index is c, use the battery characteristics and capacity values ​​in the L index as the input value of the network, and use the capacity value when the index is c+10+1 as the label to form a time series window. Then, this window rolls back one cycle each time to form a new time series window until this window rolls to the last time point; S2.2: Normalize the data. Since the types of input parameters are different and the numerical differences are large, the input parameter data needs to be normalized. The data normalization method used is min-max normalization, also known as deviation normalization, which is a linear transformation of the original data so that the result value is mapped to [0,1]. The formula is: ,in is the maximum value of a single column of feature data, is the minimum value of a single column of feature data; S2.3: The local connection characteristics of the convolution layer in the CNN network are used to extract the spatial features in the battery discharge data. The input data is output from the convolution layer and enters the pooling layer to compress and reduce the dimension of the input data to improve the quality of the data features and avoid overfitting of the model. The convolution layer extracts the features of the input variables through the convolution kernel, the ReLU function realizes nonlinear mapping, and the pooling layer downsamples and reduces the dimension of the extracted features.

[0010] Preferably, the specific method of step S3 is: S3.1: Build a neural network model for predicting the energy state of lithium-ion batteries based on a neural network model that integrates convolution and bidirectional long short-term memory networks, and determine the input layer, convolution layer, fully connected layer, and output layer of the neural network. The input layer is used to receive information about input data. The convolution layer is the core layer of the entire convolutional neural network. The weight parameters between two adjacent layers are convolved, and data features are extracted through convolution operations to reduce the number of parameters. The BiLSTM network structure consists of two LSTMs with different propagation directions. The fully connected layer integrates the features learned by the convolution layer and the pooling layer during the training process to prevent data overfitting. The output layer is a one-dimensional output obtained after the input parameters are convolved, i.e., the battery state of charge. S3.2: Input the training data set into the neural network model based on convolution and bidirectional long short-term memory network for training, with the minimization of mean square error loss function as the optimization goal to reduce the prediction error, and use the test data set for testing to obtain the ideal prediction model.

[0011] Preferably, the specific method of step S4 is: S4.1: The network layers of the pre-trained CNN-BiLSTM model are manually frozen layer by layer, and the estimation performance of the model on time series data under different numbers of frozen layers of transfer learning is observed to select the best number of frozen layers; the convolutional network is mainly used to extract and learn the common features of the data. The freezing of the CNN layer does not participate in the training and updating of parameters. Only the LSTM network layer and the fully connected layer with time series estimation are frozen and retrained; S4.2: On the CNN-BiLSTM network model that has been built based on the source domain data, share the CNN layer and low-level LSTM layer network parameters, and use the battery data of the target domain to fine-tune the high-level network parameters. Finally, complete the model migration and obtain the CNN-BiLSTM-TL network model.

[0012] Preferably, the specific method of step S5 is: The discharge data of UDDS and HWFET working conditions of 18650PF battery dataset at 25℃ are used as source domain dataset data and source domain training set. The high discharge interval 100%-80% and medium discharge interval 80%-20% data in mixed working condition Cycle4 at 25℃ are selected as target domain test datasets. The data are input into the CNN-BiLSTM-TL model after migration for testing. The mean absolute error (MAE) and root mean square error (RMSE) are used as models to evaluate the estimation accuracy of SOC and SOE. The specific evaluation formula is: , , in is the true value, is the predicted value, is the total number of samples.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention adopts a neural network model that combines convolution and bidirectional long short-term memory with transfer learning, which can quickly process time series data and can well mine the intrinsic connection between current moment data and past and future data. It consists of an input layer, a convolution layer, a fully connected layer, and an output layer, and can accurately evaluate the weight value and deviation value between the gradient update layers of the loss function.

[0014] 2. The present invention can accurately estimate SOC and SOE, optimize the battery charge and discharge control strategy, improve the battery life and efficiency, and is the key to ensuring efficient and stable operation of the battery system.

[0015] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or related technologies, the drawings required for use in the embodiments or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0017] Figure 1 This is the internal structure diagram of CNN-BiLSTM used in the present invention; Figure 2 This is a network model migration framework diagram of the present invention; Figure 3 The original data curve and the prediction curve after network training under the working condition of 100%-80% in the high discharge area of ​​the embodiment of the present invention; Figure 4 The original data curve and the prediction curve after network training under the working condition of 80%-20% of the discharge area in the embodiment of the present invention; DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0020] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0021] See also Figure 1-Figure 4 The present invention provides a technical solution: a method for predicting the state of a lithium-ion battery based on a convolution and bidirectional long short-term memory neural network model, characterized in that the method for predicting the state of a lithium-ion battery comprises the following steps: S1: Data acquisition and data preprocessing, obtaining lithium-ion battery data, including battery discharge current, voltage, SOC, SOE and other variables; S2: Use sliding window technology to construct battery discharge data under multiple temperatures and different driving cycles to form matrix data suitable for input into CNN, and normalize the data to obtain data training set and data test set; S3: Build a convolutional and bidirectional long short-term memory neural network model, namely the CNN-BiLSTM model, input the data training set into the neural network model for training, so that the neural network learns the characteristics between the data; input the data test set into the neural network model to predict the battery state of charge and health state. When the model does not achieve ideal results when testing the data test set, adjust the parameters of the neural network model until the test results are ideal; S4: Model migration: freeze the network layers of the pre-trained CNN-BiLSTM model layer by layer manually. After freezing some network layers in the CNN-BiLSTM model, fine-tune the parameters of the unfrozen network layers according to the discharge data of different types of batteries, and build a joint estimation model of SOC and SOE for different types of batteries and different discharge intervals. The neural network model of convolution and bidirectional long short-term memory network transfer learning, namely, the CNN-BiLSTM-TL network model, is obtained. S5: According to the obtained target domain CNN-BiLSTM-TL network model, the target domain data is input into the model to estimate the battery state of charge and health state. During the experiment, the CNN-BiLSTM-TL network model uses the same small sample training set as in the fine-tuning process to estimate the battery's SOC and SOE.

[0022] The specific method of step S1 is: according to the urban road cycle conditions, the voltage and energy change trends during the discharge process of the power battery tested at different temperatures, the discharge current, battery cell temperature, voltage, SOC parameter model input parameters in the data set are extracted, and the battery state of charge and energy state are predicted as target data.

[0023] The specific method of step S2 is: S2.1: Convert the data into a time series, determine the sliding time window length L to be 10 to reconstruct the data. If the current cycle index is c, use the battery characteristics and capacity values ​​in the L index as the input value of the network, and use the capacity value when the index is c+10+1 as the label to form a time series window. Then, this window rolls back one cycle each time to form a new time series window until this window rolls to the last time point; S2.2: Normalize the data. Since the types of input parameters are different and the numerical differences are large, the main solution to this problem is to normalize the input parameter data. The method used to normalize the data is min-max normalization, also known as deviation normalization, which is a linear transformation of the original data so that the result value is mapped to [0,1]. The formula is: ,in is the maximum value of a single column of feature data, is the minimum value of a single column of feature data; S2.3: The local connection characteristics of the convolution layer in the CNN network are used to extract the spatial features in the battery discharge data. The input data is output from the convolution layer and enters the pooling layer to compress and reduce the dimension of the input data to improve the quality of the data features and avoid overfitting of the model. The convolution layer extracts the features of the input variables through the convolution kernel, the ReLU function realizes nonlinear mapping, and the pooling layer downsamples and reduces the dimension of the extracted features.

[0024] The specific method of step S3 is: S3.1: Build a neural network model for predicting the energy state of lithium-ion batteries based on a neural network model that integrates convolution and bidirectional long short-term memory networks, and determine the input layer, convolution layer, fully connected layer, and output layer of the neural network. The input layer is used to receive information about input data. The convolution layer is the core layer of the entire convolutional neural network. The weight parameters between two adjacent layers are convolved, and data features are extracted through convolution operations to reduce the number of parameters. The bidirectional long short-term memory neural network (BiLSTM) model is widely used in long-term series data. The BiLSTM network structure consists of two LSTMs with different propagation directions, which can well mine the intrinsic connection between current data and past and future data. The fully connected layer is to integrate the features learned by the convolution layer and the pooling layer during the training process to prevent data overfitting. The output layer is to obtain a one-dimensional output after the input parameters are convolved, that is, the battery state of charge. S3.2: Input the training data set into the neural network model based on convolution and bidirectional long short-term memory network for training, with the minimization of mean square error loss function as the optimization goal to reduce the prediction error, and use the test data set for testing to obtain the ideal prediction model.

[0025] The specific method of step S4 is: S4.1: The network layers of the pre-trained CNN-BiLSTM model are manually frozen layer by layer, and the estimation performance of the model on time series data under different numbers of frozen layers of transfer learning is observed to select the best number of frozen layers; the convolutional network is mainly used to extract and learn the common features of the data. The freezing of the CNN layer does not participate in the training and updating of parameters. Only the LSTM network layer and the fully connected layer with time series estimation are frozen and retrained; S4.2: On the CNN-BiLSTM network model that has been built based on the source domain data, share the CNN layer and low-level LSTM layer network parameters, and use the battery data of the target domain to fine-tune the high-level network parameters. Finally, complete the model migration and obtain the CNN-BiLSTM-TL network model.

[0026] The specific method of step S5 is: The discharge data of UDDS and HWFET working conditions of 18650PF battery dataset at 25℃ are used as source domain dataset data and source domain training set. The high discharge interval 100%-80% and medium discharge interval 80%-20% data in mixed working condition Cycle4 at 25℃ are selected as target domain test datasets. The data are input into the CNN-BiLSTM-TL model after migration for testing. The mean absolute error (MAE) and root mean square error (RMSE) are used as the model to evaluate the estimation accuracy of SOC and SOE. MAE is a commonly used indicator to measure the difference between the actual value and the estimated value of the model, and is used to evaluate the degree of fit of the model on the given data. RMSE represents the difference between the actual value and the estimated value of SOC and SOE, and measures the degree of discreteness of the sample. When measuring the estimation accuracy of the model, the smaller the RMSE, the higher the estimation accuracy of the model. The specific evaluation formula is: , , in is the true value, is the predicted value, is the total number of samples. Example:

[0027] The effectiveness of the proposed method is verified using a dataset of Panasonic 18650PF lithium batteries, which are provided by the University of Wisconsin-Madison. Panasonic 18650PF is a lithium nickel cobalt aluminum chemistry battery. The dataset was collected after a series of tests under different temperature conditions and different drive cycle conditions. The dataset includes important battery parameters such as current, voltage, and battery cell temperature. The battery drive cycle test was conducted at 25°C using the Urban Dynamometer Driving Schedule (UDDS), Highway Fuel Economy Driving Schedule (HWFET), and three drive cycles: Cycle_1, Cycle_2, and Cycle_3.

[0028] Extract the discharge current, battery cell temperature, voltage, SOC and other parameter model input parameters from the urban road cycle and high-speed driving test data sets; input the obtained lithium-ion battery data into the neural network model based on the fusion of convolution and bidirectional long short-term memory network to predict the corresponding lithium-ion battery state of charge and health state change trend, and predict the battery state of charge and energy state according to the battery state prediction curve of the neural network model based on the fusion of convolution and bidirectional long short-term memory network.

[0029] The network structure and parameter settings of the neural network model that integrates convolution and gated recurrent units are shown in Table 1 below: Table 1 Neural network model network structure and parameter settings

[0030] The original data curve and prediction curve after network training are as follows Figure 3 and 4 As shown: The simulation results are shown in Table 2 below: Table 2 Simulation experiment results

[0031] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting lithium-ion battery status based on a neural network model of convolution and bidirectional long short-term memory, characterized in that: The lithium-ion battery state prediction method comprises the following steps: S1: Data acquisition and data preprocessing, obtaining lithium-ion battery data, including battery discharge current, voltage, SOC and SOE variables; S2: Use sliding window technology to construct battery discharge data under multiple temperatures and different driving cycles to form matrix data suitable for input into CNN, and normalize the data to obtain data training set and data test set; S3: Build a convolutional and bidirectional long short-term memory neural network model, namely the CNN-BiLSTM model, input the data training set into the neural network model for training, so that the neural network learns the characteristics between the data; input the data test set into the neural network model to predict the battery state of charge and health state. When the model does not achieve ideal results when testing the data test set, adjust the parameters of the neural network model until the test results are ideal; S4: Model migration: freeze the network layers of the pre-trained CNN-BiLSTM model layer by layer manually. After freezing some network layers in the CNN-BiLSTM model, fine-tune the parameters of the unfrozen network layers according to the discharge data of different types of batteries, and build a joint estimation model of SOC and SOE for different types of batteries and different discharge intervals. The neural network model of convolution and bidirectional long short-term memory network transfer learning, namely, the CNN-BiLSTM-TL network model, is obtained. S5: According to the obtained target domain CNN-BiLSTM-TL network model, the target domain data is input into the model to estimate the battery state of charge and health state. During the experiment, the CNN-BiLSTM-TL network model uses the same small sample training set as in the fine-tuning process to estimate the battery's SOC and SOE.

2. The method for predicting lithium-ion battery status based on a convolution and bidirectional long short-term memory neural network model according to claim 1, characterized in that: The specific method of step S1 is: according to the urban road cycle conditions, the voltage and energy change trends during the discharge process of the power battery tested at different temperatures are extracted, the discharge current, battery cell temperature, voltage, SOC parameter model input parameters in the data set are extracted, and the battery state of charge and energy state are predicted as target data.

3. The method for predicting lithium-ion battery status based on a neural network model of convolution and bidirectional long short-term memory according to claim 1, characterized in that: The specific method of step S2 is: S2.1: Convert the data into a time series, determine the sliding time window length L to be 10 to reconstruct the data. If the current cycle index is c, use the battery characteristics and capacity values ​​in the L index as the input value of the network, and use the capacity value when the index is c+10+1 as the label to form a time series window. Then, this window rolls back one cycle each time to form a new time series window until this window rolls to the last time point; S2.2: Normalize the data. Since the types of input parameters are different and the numerical differences are large, the input parameter data needs to be normalized. The data normalization method used is min-max normalization, also known as deviation normalization, which is a linear transformation of the original data so that the result value is mapped to [0,1]. The formula is: ,in is the maximum value of a single column of feature data, is the minimum value of a single column of feature data; S2.3: The local connection characteristics of the convolution layer in the CNN network are used to extract the spatial features in the battery discharge data. The input data is output from the convolution layer and enters the pooling layer to compress and reduce the dimension of the input data to improve the quality of the data features and avoid overfitting of the model. The convolution layer extracts the features of the input variables through the convolution kernel, the ReLU function realizes nonlinear mapping, and the pooling layer downsamples and reduces the dimension of the extracted features.

4. The method for predicting lithium-ion battery status based on a convolution and bidirectional long short-term memory neural network model according to claim 1, characterized in that: The specific method of step S3 is: S3.1: Build a neural network model for predicting the energy state of lithium-ion batteries based on a neural network model that integrates convolution and bidirectional long short-term memory networks, and determine the input layer, convolution layer, fully connected layer, and output layer of the neural network. The input layer is used to receive information about input data. The convolution layer is the core layer of the entire convolutional neural network. The weight parameters between two adjacent layers are convolved, and data features are extracted through convolution operations to reduce the number of parameters. The BiLSTM network structure consists of two LSTMs with different propagation directions. The fully connected layer integrates the features learned by the convolution layer and the pooling layer during the training process to prevent data overfitting. The output layer is a one-dimensional output obtained after the input parameters are convolved, i.e., the battery state of charge. S3.2: Input the training data set into the neural network model based on convolution and bidirectional long short-term memory network for training, with the minimization of mean square error loss function as the optimization goal to reduce the prediction error, and use the test data set for testing to obtain the ideal prediction model.

5. The method for predicting lithium-ion battery status based on a convolution and bidirectional long short-term memory neural network model according to claim 1, characterized in that: The specific method of step S4 is: S4.1: The network layers of the pre-trained CNN-BiLSTM model are manually frozen layer by layer, and the estimation performance of the model on time series data under different numbers of frozen layers of transfer learning is observed to select the best number of frozen layers; the convolutional network is mainly used to extract and learn the common features of the data. The freezing of the CNN layer does not participate in the training and updating of parameters. Only the LSTM network layer and the fully connected layer with time series estimation are frozen and retrained; S4.2: On the CNN-BiLSTM network model that has been built based on the source domain data, share the CNN layer and low-level LSTM layer network parameters, and use the battery data of the target domain to fine-tune the high-level network parameters. Finally, complete the model migration and obtain the CNN-BiLSTM-TL network model.

6. The method for predicting lithium-ion battery status based on a convolution and bidirectional long short-term memory neural network model according to claim 1, characterized in that: The specific method of step S5 is: The discharge data of UDDS and HWFET working conditions of 18650PF battery dataset at 25℃ are used as source domain dataset data and source domain training set. The high discharge interval 100%-80% and medium discharge interval 80%-20% data in mixed working condition Cycle4 at 25℃ are selected as target domain test datasets. The data are input into the CNN-BiLSTM-TL model after migration for testing. The mean absolute error (MAE) and root mean square error (RMSE) are used as models to evaluate the estimation accuracy of SOC and SOE. The specific evaluation formula is: ， ， in is the true value, is the predicted value, is the total number of samples.

Images